Predictive Modelling of Advertising Awareness

1
Predictive Modelling of Advertising Awareness
2
A motivating example
3
Key Questions

• How do you know you are using your media budget
  to maximum effect
• Which executions are working best?
• Are some wearing out?
• is our sceduling right?
• What is the best flighting strategy?
• Does this lead to an increase in market share?

4
How advertisng is modelled
5
How advertisng is modelled...
6
How advertisng is modelled...
New
7
How advertisng is modelled...
8
Adstock Modelling

• Poor correlation with Ad recall and TARPS
• Much better correlation with Adstock
• Adstock gives TARPS memory
• So Recall and Adstock are comparable
• Ad recallt Legacy Impact . Adstockt
• Legacy long term memory
• Decay rate at which people forget
• Impact rate of return of recall/100 TARPS

9
How is Adstock modelled

• . Adstockt dTarpst (1-d) . Adstockt-1
• where d decay rate usually about 10 or less
• Initial value taken to be Adstock1 dTarps1
• Exponentially smoothes Tarps so they become
  continuous
• Now have a memory component like recall

10
Motivating example revisited.How good is the
model?
400
350
300
250
ECT
200
TARPs
150
100
50
0
9/7/00
6/8/00
3/9/00
28/5/00
11/6/00
25/6/00
23/7/00
20/8/00
17/9/00
1/10/00
15/10/00
29/10/00
12/11/00
26/11/00
date
Modelled NETT ECT
NETT ECT
TARPS
11
Motivating example
Impact Indices
4.5
4.0
Ad A
3.5
Ad B
3.0
Impact
Ad C
2.5
Ad D
2.0
Ad E
1.5
Average
1.0
2/7/00
3/9/00
30/4/00
21/5/00
11/6/00
23/7/00
13/8/00
24/9/00
5/11/00
15/10/00
Ads A E return the best value
12
Future Media Spend - some scenarios
13
Proposed spend until June 2001(1500 TARPS in 10
weeks)

• 12 low builds slowly to 21 ECT
• Average ECT 19 after February

14
Alternative Spend Until June(Same Budget)

• Average ECT 21
• Burst and hold Strategy
• ECT higher longer - less variation

15
Whats been happening with this campaign lately?
ECT showing immediate increase following re-start
of campaign
16
Modelled data and prediction

• Model adjusted to account for actual ECT and
  current spend will see a return to average ECT of
  approximately 20-25

17
Dynamic Adstock Modelling

• Impact can be evaluated on a weekly basis to see
  if it changes with time. This can indicate when
• An ad is wearing out
• Or if some other external factor is influencing
  awareness e.g.
• Better flight / channelling
• Increased clutter in the market

18
Ad A - Impact (return/100 TARPs)
Ad wearing out with time.
19
Ad. B - Impact ( return/100 TARPs)
Same spend -different channels.
20
Key Learnings

• Thresholds of under/overspending exist
• Avoid 15 second executions
• Do not run multiple creative executions
• SOV is critical
• As executions may appear to be wearing out when
  in fact competition consumers ear has increased
• Burst and maintain strategy works best in the
  markets analysed to date

21
Advertising modelling can be used to

• Diagnose the effectiveness and current health of
  each execution
• Predict potential future scenarios
• find the optimal media expenditure strategy

22
The Relationship to Market Share

• Getting awareness up is first base
• it doesnt necessarily result in increased share
• however, chances are that the client will notice
  the effects when the ad is not on
• In other words, it is a composite of optimal
  spending on advertising and what is happening in
  terms of distribution/sales and service.
• Or -its a bloody hard problem!!!

23
51 of Brand share explained by what we measure
Model Fit
33 of model fit due to adstock alone
10
9
Brand Share
8
7
Execution A
Execution B
0
20
40
60
80
100
Date
24
A Market Share Model

• BRANDSHARE
• 5.830053 initial
• -2.16682WINTER Opposition dumps!
• 0.547SOVLOTS SOV gt40
• 0.031Adstock
• 0.052AdsExA Execution A lifts Share
• -0.0006AdsExA2 Overspend on Ex A